[PATCH 3/8] drm/amdgpu: Implement mmap as GEM object function
Thomas Zimmermann
tzimmermann at suse.de
Tue Apr 6 10:38:09 UTC 2021
Hi
Am 06.04.21 um 11:35 schrieb Christian König:
> Am 06.04.21 um 11:08 schrieb Thomas Zimmermann:
>> Moving the driver-specific mmap code into a GEM object function allows
>> for using DRM helpers for various mmap callbacks.
>>
>> This change resolves several inconsistencies between regular mmap and
>> prime-based mmap. The vm_ops field in vma is now set for all mmap'ed
>> areas. Previously it way only set for regular mmap calls, prime-based
>> mmap used TTM's default vm_ops. The check for kfd_bo has been taken
>> from amdgpu_verify_access(), which is not called any longer and has
>> been removed.
>>
>> As a side effect, amdgpu_ttm_vm_ops and amdgpu_ttm_fault() are now
>> implemented in amdgpu's GEM code.
>>
>> Signed-off-by: Thomas Zimmermann <tzimmermann at suse.de>
>> ---
>> drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c | 46 -------------
>> drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h | 2 -
>> drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 4 +-
>> drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 64 +++++++++++++++++++
>> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 71 ---------------------
>> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 1 -
>> 6 files changed, 66 insertions(+), 122 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c
>> index e0c4f7c7f1b9..19c5ab08d9ec 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c
>> @@ -42,52 +42,6 @@
>> #include <linux/pci-p2pdma.h>
>> #include <linux/pm_runtime.h>
>> -/**
>> - * amdgpu_gem_prime_mmap - &drm_driver.gem_prime_mmap implementation
>> - * @obj: GEM BO
>> - * @vma: Virtual memory area
>> - *
>> - * Sets up a userspace mapping of the BO's memory in the given
>> - * virtual memory area.
>> - *
>> - * Returns:
>> - * 0 on success or a negative error code on failure.
>> - */
>> -int amdgpu_gem_prime_mmap(struct drm_gem_object *obj,
>> - struct vm_area_struct *vma)
>> -{
>> - struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
>> - struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
>> - unsigned asize = amdgpu_bo_size(bo);
>> - int ret;
>> -
>> - if (!vma->vm_file)
>> - return -ENODEV;
>> -
>> - if (adev == NULL)
>> - return -ENODEV;
>> -
>> - /* Check for valid size. */
>> - if (asize < vma->vm_end - vma->vm_start)
>> - return -EINVAL;
>> -
>> - if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm) ||
>> - (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)) {
>> - return -EPERM;
>> - }
>> - vma->vm_pgoff += amdgpu_bo_mmap_offset(bo) >> PAGE_SHIFT;
>> -
>> - /* prime mmap does not need to check access, so allow here */
>> - ret = drm_vma_node_allow(&obj->vma_node,
>> vma->vm_file->private_data);
>> - if (ret)
>> - return ret;
>> -
>> - ret = ttm_bo_mmap(vma->vm_file, vma, &adev->mman.bdev);
>> - drm_vma_node_revoke(&obj->vma_node, vma->vm_file->private_data);
>> -
>> - return ret;
>> -}
>> -
>> static int
>> __dma_resv_make_exclusive(struct dma_resv *obj)
>> {
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h
>> index 39b5b9616fd8..3e93b9b407a9 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h
>> @@ -31,8 +31,6 @@ struct drm_gem_object
>> *amdgpu_gem_prime_import(struct drm_device *dev,
>> struct dma_buf *dma_buf);
>> bool amdgpu_dmabuf_is_xgmi_accessible(struct amdgpu_device *adev,
>> struct amdgpu_bo *bo);
>> -int amdgpu_gem_prime_mmap(struct drm_gem_object *obj,
>> - struct vm_area_struct *vma);
>> extern const struct dma_buf_ops amdgpu_dmabuf_ops;
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>> index 76f48f79c70b..e96d2758f4bb 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>> @@ -1656,7 +1656,7 @@ static const struct file_operations
>> amdgpu_driver_kms_fops = {
>> .flush = amdgpu_flush,
>> .release = drm_release,
>> .unlocked_ioctl = amdgpu_drm_ioctl,
>> - .mmap = amdgpu_mmap,
>> + .mmap = drm_gem_mmap,
>> .poll = drm_poll,
>> .read = drm_read,
>> #ifdef CONFIG_COMPAT
>> @@ -1719,7 +1719,7 @@ static const struct drm_driver amdgpu_kms_driver
>> = {
>> .prime_handle_to_fd = drm_gem_prime_handle_to_fd,
>> .prime_fd_to_handle = drm_gem_prime_fd_to_handle,
>> .gem_prime_import = amdgpu_gem_prime_import,
>> - .gem_prime_mmap = amdgpu_gem_prime_mmap,
>> + .gem_prime_mmap = drm_gem_prime_mmap,
>> .name = DRIVER_NAME,
>> .desc = DRIVER_DESC,
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
>> index fb7171e5507c..fe93faad05f2 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
>> @@ -41,6 +41,36 @@
>> static const struct drm_gem_object_funcs amdgpu_gem_object_funcs;
>> +static vm_fault_t amdgpu_ttm_fault(struct vm_fault *vmf)
>
> Please name that function amdgpu_gem_fault or amdgpu_gem_object_fault
>
>> +{
>> + struct ttm_buffer_object *bo = vmf->vma->vm_private_data;
>> + vm_fault_t ret;
>> +
>> + ret = ttm_bo_vm_reserve(bo, vmf);
>> + if (ret)
>> + return ret;
>> +
>> + ret = amdgpu_bo_fault_reserve_notify(bo);
>> + if (ret)
>> + goto unlock;
>> +
>> + ret = ttm_bo_vm_fault_reserved(vmf, vmf->vma->vm_page_prot,
>> + TTM_BO_VM_NUM_PREFAULT, 1);
>> + if (ret == VM_FAULT_RETRY && !(vmf->flags &
>> FAULT_FLAG_RETRY_NOWAIT))
>> + return ret;
>> +
>> +unlock:
>> + dma_resv_unlock(bo->base.resv);
>> + return ret;
>> +}
>> +
>> +static const struct vm_operations_struct amdgpu_ttm_vm_ops = {
>> + .fault = amdgpu_ttm_fault,
>> + .open = ttm_bo_vm_open,
>> + .close = ttm_bo_vm_close,
>> + .access = ttm_bo_vm_access
>> +};
>> +
>> static void amdgpu_gem_object_free(struct drm_gem_object *gobj)
>> {
>> struct amdgpu_bo *robj = gem_to_amdgpu_bo(gobj);
>> @@ -201,6 +231,38 @@ static void amdgpu_gem_object_close(struct
>> drm_gem_object *obj,
>> ttm_eu_backoff_reservation(&ticket, &list);
>> }
>> +static int amdgpu_gem_prime_mmap(struct drm_gem_object *obj, struct
>> vm_area_struct *vma)
>> +{
>> + struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
>> + struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
>> + unsigned long asize = amdgpu_bo_size(bo);
>> +
>> + if (!vma->vm_file)
>> + return -ENODEV;
>> +
>> + if (!adev)
>> + return -ENODEV;
>> +
>> + /* Check for valid size. */
>> + if (asize < vma->vm_end - vma->vm_start)
>> + return -EINVAL;
>
>> +
>> + /*
>> + * Don't verify access for KFD BOs. They don't have a GEM
>> + * object associated with them.
>> + */
>> + if (bo->kfd_bo)
>> + goto out;
>
> Who does the access verification now?
This is somewhat confusing.
I took this check as-is, including the comment, from amdgpu's
verify_access function. The verify_access function was called by
ttm_bo_mmap. It returned 0 and ttm_bo_mmap did the mapping.
>
> Christian.
>
>> +
>> + if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm) ||
>> + (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)) {
>> + return -EPERM;
>> + }
>> +
>> +out:
>> + return drm_gem_ttm_mmap(obj, vma);
>> +}
>> +
>> static const struct drm_gem_object_funcs amdgpu_gem_object_funcs = {
>> .free = amdgpu_gem_object_free,
>> .open = amdgpu_gem_object_open,
>> @@ -208,6 +270,8 @@ static const struct drm_gem_object_funcs
>> amdgpu_gem_object_funcs = {
>> .export = amdgpu_gem_prime_export,
>> .vmap = drm_gem_ttm_vmap,
>> .vunmap = drm_gem_ttm_vunmap,
>> + .mmap = amdgpu_gem_prime_mmap,
>> + .vm_ops = &amdgpu_ttm_vm_ops,
>> };
>> /*
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> index 1c6131489a85..d9de91a517c6 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> @@ -152,32 +152,6 @@ static void amdgpu_evict_flags(struct
>> ttm_buffer_object *bo,
>> *placement = abo->placement;
>> }
>> -/**
>> - * amdgpu_verify_access - Verify access for a mmap call
>> - *
>> - * @bo: The buffer object to map
>> - * @filp: The file pointer from the process performing the mmap
>> - *
>> - * This is called by ttm_bo_mmap() to verify whether a process
>> - * has the right to mmap a BO to their process space.
>> - */
>> -static int amdgpu_verify_access(struct ttm_buffer_object *bo, struct
>> file *filp)
>> -{
>> - struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo);
>> -
>> - /*
>> - * Don't verify access for KFD BOs. They don't have a GEM
>> - * object associated with them.
>> - */
>> - if (abo->kfd_bo)
>> - return 0;
>> -
>> - if (amdgpu_ttm_tt_get_usermm(bo->ttm))
>> - return -EPERM;
>> - return drm_vma_node_verify_access(&abo->tbo.base.vma_node,
>> - filp->private_data);
>> -}
Here's the orignal verification code. It gives a free pass to KFD.
>> -
>> /**
>> * amdgpu_ttm_map_buffer - Map memory into the GART windows
>> * @bo: buffer object to map
>> @@ -1531,7 +1505,6 @@ static struct ttm_device_funcs amdgpu_bo_driver = {
>> .eviction_valuable = amdgpu_ttm_bo_eviction_valuable,
>> .evict_flags = &amdgpu_evict_flags,
>> .move = &amdgpu_bo_move,
>> - .verify_access = &amdgpu_verify_access,
>> .delete_mem_notify = &amdgpu_bo_delete_mem_notify,
>> .release_notify = &amdgpu_bo_release_notify,
>> .io_mem_reserve = &amdgpu_ttm_io_mem_reserve,
>> @@ -1906,50 +1879,6 @@ void amdgpu_ttm_set_buffer_funcs_status(struct
>> amdgpu_device *adev, bool enable)
>> adev->mman.buffer_funcs_enabled = enable;
>> }
>> -static vm_fault_t amdgpu_ttm_fault(struct vm_fault *vmf)
>> -{
>> - struct ttm_buffer_object *bo = vmf->vma->vm_private_data;
>> - vm_fault_t ret;
>> -
>> - ret = ttm_bo_vm_reserve(bo, vmf);
>> - if (ret)
>> - return ret;
>> -
>> - ret = amdgpu_bo_fault_reserve_notify(bo);
>> - if (ret)
>> - goto unlock;
>> -
>> - ret = ttm_bo_vm_fault_reserved(vmf, vmf->vma->vm_page_prot,
>> - TTM_BO_VM_NUM_PREFAULT, 1);
>> - if (ret == VM_FAULT_RETRY && !(vmf->flags &
>> FAULT_FLAG_RETRY_NOWAIT))
>> - return ret;
>> -
>> -unlock:
>> - dma_resv_unlock(bo->base.resv);
>> - return ret;
>> -}
>> -
>> -static const struct vm_operations_struct amdgpu_ttm_vm_ops = {
>> - .fault = amdgpu_ttm_fault,
>> - .open = ttm_bo_vm_open,
>> - .close = ttm_bo_vm_close,
>> - .access = ttm_bo_vm_access
>> -};
>> -
>> -int amdgpu_mmap(struct file *filp, struct vm_area_struct *vma)
>> -{
>> - struct drm_file *file_priv = filp->private_data;
>> - struct amdgpu_device *adev = drm_to_adev(file_priv->minor->dev);
>> - int r;
>> -
>> - r = ttm_bo_mmap(filp, vma, &adev->mman.bdev);
>> - if (unlikely(r != 0))
>> - return r;
>> -
>> - vma->vm_ops = &amdgpu_ttm_vm_ops;
>> - return 0;
>> -}
And this was the mmap callback in struct file_operations. It calls
ttm_bo_mmap(), which skips verification for KFD BOs. To the best of my
knowledge, there was no additional verification for these KFD BOs.
The original code in amdgpu_gem_prime_mmap() did seom verification, but
didn't handle KFD specially. I guess, PRIME needs GEM and KFD BOs
wouldn't quailify.
In the end I went with the semantics I found in amdgpu_mmap() and
handled KFD specially. Let me know if this requires to be changed.
Best regards
Thomas
>> -
>> int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset,
>> uint64_t dst_offset, uint32_t byte_count,
>> struct dma_resv *resv,
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>> index dec0db8b0b13..6e51faad7371 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>> @@ -146,7 +146,6 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo,
>> struct dma_resv *resv,
>> struct dma_fence **fence);
>> -int amdgpu_mmap(struct file *filp, struct vm_area_struct *vma);
>> int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo);
>> int amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo);
>> uint64_t amdgpu_ttm_domain_start(struct amdgpu_device *adev,
>> uint32_t type);
>
> _______________________________________________
> dri-devel mailing list
> dri-devel at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel
--
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature
Type: application/pgp-signature
Size: 840 bytes
Desc: OpenPGP digital signature
URL: <https://lists.freedesktop.org/archives/dri-devel/attachments/20210406/fd16485b/attachment-0001.sig>
More information about the dri-devel
mailing list